38
Potential energy surface and molecular dynamics simulations Yaroslava G. Yingling
Potential energy surface and
molecular dynamics simulations
Yaroslava G. Yingling
Potential Energy Surface• A potential energy surface is a mathematical function that gives the energy
of a molecule as a function of its geometry. • Does not change if it translated or rotated in space• Depends on a molecule’s internal coordinates • Internal coordinates may be represented by simple stretch, bend, torsion, or normal
model, etc.
• Quantum Mechanics provides an energy function which can be exact in principle and works for any molecule. In practice, approximate quantum methods are used due to computational expense.• Molecular Mechanics provides this energy as a function of stretches,
bends, torsions, etc. This is an approximate model that breaks down in some situations (e.g., breaking bonds). Only works when parameters are available.
Molecular Mechanics• Simplest type of calculation
• Used when systems are very large and approaches that are more accurate become too costly (in time and memory)
• Molecular Mechanics is a method to calculate the structure and energy of molecules based on nuclear motions.
• electrons are not considered explicitly• will find optimum distribution once position of nuclei are known• Born-Oppenheimer approximation of Shrödinger equation
➢ nuclei are heavier and move slower than electrons ➢ nuclear motions (vibrations, rotations) can be studied separately➢ electrons move fast enough to adjust to any nuclei movement
• Basic assumptions:• Each atom is represented by a particle (m, R)• Chemical bond represented by a spring• Potential energy function described intramolecular phenomena and rely on empirically
derived parameters
Classical description of interatomic potential• Electrons are not present explicitly, they are introduced through the potential
energy surface that is a function of atomic positions only (Born-Oppenheimer approximation).
• The potential energy surface, in turn, is approximated by an analytic function that gives the potential energy U as a function of coordinates. Forces are obtained as the gradient of a potential energy function
• The choice of a potential function that approximates the actual (unknown)
solution of the Schrödinger equation is a difficult task.
• Design of the potential function and choice of the parameters is often based on
fitting to available experimental data (e.g. equilibrium geometry of stable phases,
cohesive energy, elastic moduli, vibrational frequencies, temperatures of the
phase transitions, etc.).
Molecular Mechanics• Interactions between atoms (Potential Energy Function) are represented by functions of distance, angle or dihedral
• Collection of empirical parameters and potential functions is known as a force field.
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
Molecular Dynamics• Potential function for MD1,2
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
1. Levitt M. Hirshberg M. Sharon R. Daggett V. Comp. Phys. Comm. (1995) 91: 215-231
2. Levitt M. et al. J. Phys. Chem. B (1997) 101: 5051-5061
Molecular Dynamics• Potential function for MD1,2
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
1. Levitt M. Hirshberg M. Sharon R. Daggett V. Comp. Phys. Comm. (1995) 91: 215-231
2. Levitt M. et al. J. Phys. Chem. B (1997) 101: 5051-5061
b0
Molecular Dynamics• Potential function for MD1,2
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
1. Levitt M. Hirshberg M. Sharon R. Daggett V. Comp. Phys. Comm. (1995) 91: 215-231
2. Levitt M. et al. J. Phys. Chem. B (1997) 101: 5051-5061
θ0
Molecular Dynamics• Potential function for MD1,2
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
1. Levitt M. Hirshberg M. Sharon R. Daggett V. Comp. Phys. Comm. (1995) 91 215-231
2. Levitt M. et al. J. Phys. Chem. B (1997) 101:25 5051-5061
Φ0
Molecular Dynamics• Potential function for MD1,2
U = Bond + Angle + Dihedral + van der Waals + Electrostatic
1. Levitt M. Hirshberg M. Sharon R. Daggett V. Comp. Phys. Comm. (1995) 91: 215-231
2. Levitt M. et al. J. Phys. Chem. B (1997) 101: 5051-5061
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
To a large degree, soft materials structure is dependent on non-bonded atomic interactions
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
+ -
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
+ +
December 14, 2017 James S Peerless17
Partial Charges
Dihedral
Bond
Angle
Static Partial Charges (qi,qj) are typically parameterized for each molecule prior to simulation
As a result, MD practitioners use ab initio or semi-empirical quantum mechanical calculations to estimate partial charges:• RESP: Restrained ElectroStatic Potential mapping
• Uses ab initio calculations to construct electrostatic potential map (MEP)
• Least squares algorithm derives atom centered charges that best reproduces MEP
• Charges then assigned to reduce error between ab initio electrostatic potential map and that constructed from charge assignment – “buried” atoms restrained to minimize charge
• Results depend on ab initio QM method used to construct ESP
• Semi-Emperical (AM1-BCC, CM1, CM2, Mulliken):• Semi-empirical partial charges assigned based on Core
Repulsion Function (CRF) and orbital structures• Bond Charge Correction (BCC) attempts to distribut charge
based on bonds to adjacent atoms
December 14, 2017 James S Peerless18
Test Case: Acetonitrile (ACN)• Polar aprotic solvent – commonly used in
organic synthesis
• Polarity affects solvent properties
• 4 Symmetrically unique atoms• 3 DoF for partial charge• 1 buried atom (C2)
• Subject of many previous simulation studies in which partial charges were calculated dating back to 1983
Atom Korvega, 2017B-LYP&TM/90Ry
Korvega, 2017B-LYP&TM + vdW/90Ry
Caleman, 2012OPLS/AA
Caleman, 2012GAFF
Grabuleda, 20006-31G*
Grabuleda, 20006-311+G*
Nikitin,2007
Cabaleiro-Lago,1997 HF fitted
Cabaleiro-Lago, 1997MP2 fitted
RED Server,HF/6-31G*
AM1-BCC
N −0.475 −0.475 −0.56 −0.5168 -0.490 -0.532 −0.5126 −0.532 −0.494 -0.4936 -0.3758
C1 0.305 0.305 0.46 0.4484 0.382 0.481 0.4917 0.481 0.475 0.3924 0.2087
C2 0.185 0.182 −0.08 −0.4008 -0.2376 -0.479 −0.5503 −0.479 −0.552 -0.2543 -0.045
H −0.005 −0.004 0.06 0.1564 0.115 0.177 0.1904 0.117 0.190 0.1185 0.0707
C1
C2
NH
Dipole Moment
Importance of partial charges
Example: polyvinyl chloride (PVC):
• 2nd/3rd most common plastic worldwide
• Applications:
o Automotive (dashboard & underbody coating)
o Construction (flooring & window blinds)
o Electronics (cable & wire insulation)
o Medical (IV bags, gloves, & tubing)…
PVC: Koenhen, D. M. & Smolders, C. A. (1975). Journal of Applied Polymer Science, 19: 1163-1179
HSP
(M
Pa1
/2)
• Partial Charges (qi,qj) are parameterized for each molecule prior to simulation• Do not change during simulation – Born-Oppenheimer approximation• We use ab initio or semi-empirical calculations to estimate partial charges
Molecular Dynamics• Non-bonded components of potential function
Unb = van der Waals + Electrostatic
NOTE:Sum over all pairs of N atoms
2
1− NN
N is often between 5x105 to 5x106
For 5x105 that is 1.25x1011 pairs
THAT IS A LOT OF POSSIBLE PAIRS!
Force Field used to calculate the energy and geometry of a molecule.
• Collection of atom types (to define the atoms in a molecule), parameters (for bond lengths,
bond angles, etc.) and equations (to calculate the energy of a molecule)
• In a force field, a given element may have several atom types.
➢ For example, phenylalanine contains both sp3-hybridized carbons and aromatic carbons.
➢ sp3-Hybridized carbons have a tetrahedral bonding geometry
➢ aromatic carbons have a trigonal bonding geometry.
➢ C-C bond in the ethyl group differs from a C-C bond in the phenyl ring
➢ C-C bond between the phenyl ring and the ethyl group differs from all other C-C
bonds in ethylbenzene. The force field contains parameters for these different
types of bonds.
Force Field
• Molecular Mechanics or force-field methods use classical type models to predict the energy of a molecule as a function of its conformation. This allows predictions of
• Equilibrium geometries and transition states
• Relative energies between conformers or between different molecules
• Molecular mechanics can be used to supply the potential energy for molecular dynamics computations on large molecules.
Local versus Global minimum problem
• Structural landscape is filled
with peaks and valleys.
• Minimization protocol always
moves “down hill”.
• No means to “see” the overall
structural landscape
• No means to pass through
higher intermediate structures
to get to a lower minima.
The initial structure
determines the results of the
minimization!
Molecular geometry minimization
The molecular potential U depends on two types of variables:
Potential energy gradient g(Q), a vector with 3N components:
The necessary condition for a minimum is that the function
gradient is zero:
Where xi denote atomic Cartesian coordinates and N is the number
of atoms
or
One measure of the distance from a stationary point is the rms gradient:
The sufficient condition for a minimum is that the second
derivative matrix is positive definite, i.e. for any 3N-dimensional
vector u:A simpler operational definition of this property is that all
eigenvalues of F are positive at a minimum. The second
derivative matrix is denoted by F in molecular mechanics and H in
mathematics, and is defined as:
Molecular Minimization: First-order derivative methods
Steepest Descent Method
⧫ 1.) Evaluate the sum of all forces on the system (first derivative
of potential energy functions)
⧫2.) Move in the direction of the force until potential energy stops
decreasing
⧫3.) Turn 90° and return to step 2
⧫ not efficient, but good for initial distorted structures
⧫ may be very slow near a solution
Conjugate Gradient (Powell)
⧫1.) Evaluate the sum of all forces on the
system (first derivative of potential energy
functions)
⧫2.) Move in the direction of the force until
potential energy stops decreasing
⧫3.) Return to step 1
When to stop:
• After a predefined energy minimum has been
reached
• For example, < 1.0 kJ / mol
• After a predefined number of steps
• For example, after 1000 orthogonal steps
Initial Refinement
(Av. gradient < 1 kcal
Å-2)
Stringent
Minimization
(Av. gradient < 0.1
kcal Å-2)
MethodCPU time
(s)
# of
iterations
CPU time
(s)
# of
iterations
Steepest
descents67 98 1405 1893
Conjugate
gradients149 213 257 367
Inherent Problem of Energy Minimization
• Only finds local minima
• No method available can find the global minimum from any starting point
Classical Mechanics• Instead of using Quantum mechanics, we can use classical Newtonian
mechanics to model our system.
• Classical mechanics is the study of the motion of bodies in accordance with the general principles first enunciated by Sir Isaac Newton in his Philosophiae Naturalis Principia Mathematica (1687).
• Classical mechanics is concerned with the set of physical laws describing the motion of bodies under the action of a system of forces.
• Classical mechanics is the foundation upon which all other branches of Physics are built. It has many important applications in many areas of science:
• Astronomy (motion of stars and planets)
• Molecular and nuclear physics (collisions of atomic and subatomic particles)
• Geology (e.g., the propagation of seismic waves)
• Engineering (eg structures of bridges and buildings)
Sir Isaac Newton
1642-1727
Techniques:
• Molecular Mechanics
• Molecular Dynamics
• Dissipative Particle Dynamics
r F = m
r a
Molecular Dynamics simulations
• MD is the solution of the classical equations of motion for atoms and molecules to obtain the time evolution of the system.
• Applied to many-particle systems - a general analytical solution not possible. Must resort to numerical methods and computers
• Maxwell-Boltzmann averaging process for thermodynamic properties (time averaging).
MD simulations provide a molecular level picture of structure function relationships.
Molecular dynamics simulations
Solvation models
• Explicit solvent (any solvent)
• Implicit or continuum solvation
)]cos(1[2
)()(612
dihedralsangles
2
bonds
2
point
+
−+−++−+−=
ji ij
ji
ij
ij
ij
ijneqeqr
R
R
B
R
An
VKrrKE
1 1 1(1 )
2 2
i j i jelec
GBi j ijij w ij
q q q qE
r f
− −
2 2 1/2[ exp( / 4 )]GB
ij i j ij i jf r R R r R R= + −
Minimal input
Different Ensembles• Vary the independent variables
• Useful for describing open and closed physical systems
• Variants:
• NVE: microcanonical
• NVT: canonical
• NPT: isobaric-isothermal
• NσT: constant stress
• μVT: grand canonical
• NPH: isenthalpic-isothermal
NVE
NVT
NPT
NPH
μVTNσT
μVE
The HIV capsid is large, containing
about 1,300 proteins with altogether 4 million atoms
Simulations of 64 million atoms for over 1 micro second allow to conduct a comprehensive study of the physical properties of the entire HIV capsid including electrostatic potential, allatom normal modes, as well as the effects of the solvent (ions and water) on the capsid. The results from the simulations reveal critical details of the capsid protein with important implications for assembly, uncoatingand nuclear import.
It took two years on a supercomputer to simulate 1.2 microseconds in the life of the HIV capsid, a protein cage that shuttles the HIV virus to the nucleus of a human cell. The 64-million-atom simulation offers new insights into how the virus senses its environment and completes its infective cycle.
• MD is a deterministic technique: given initial positions and velocities, the evolution of the system in time is, in principle, completely determined (in practice, accumulation of integration and computational errors would introduce some uncertainty into the MD output).
• MD can be also used as a statistical mechanics method: it generates a set of configurations that are distributed according to statistical distribution functions. In many cases we are not interested in trajectories of individual atoms, we are interested in macroscopic properties of the material. In these cases MD information can be averaged over all the atoms in the system and over time of the simulation to obtain thermodynamic parameters.
Atomistic MD simulationsAdvantages:
• Excellent for studies of processes and properties of soft or/and hard materials
• Observe time evolution of the system
• Formation of non-bonded interactions
• Allowing the derivation of kinetic and thermodynamic properties
Limitations:
• No quantum effects: changes in chemical bonding, catalytic cleavage, presence of intermediates, light induced charge transfer, tunnelling of protons and electrons
• Workaround: ab initio (first principles) MD – not practical for large systems, base on ground state DFT
• Reactive force fields
• Quality of the force field - The results of simulations will be realistic only if the potential energy function mimics the forces experienced by the ‘real’ atoms.
• Time and size limitations - The desired length of simulations also places limits on increasing the size of the problem
Commonly used software
FundingNSF DMR; NSF CBET; NSF CMMI-1150682 (CAREER); NSF DMR-1121107 (MRSEC); NSF DMR-1205670; NSF ENG-0967559;DOE BES-EFRC DE-SC0001090 (EFRC); Eastman Chemical; BentTech Inc; RENCI
Group Members
Current members:Research Scientist & Postdocs: Dr. Abhishek Singh; Dr. Albert Kwansa; Dr. Nan Li; Dr. Hoshin KimGraduate students:Thomas Deaton, James Peerless, Yuxin Xie, Matthew Manning, Thomas Oweida, Nina Milliken
Alumni:Postdocs: Dr. Sanket Deshmukh, Dr. Stacy Snyder, Dr. Andriy Semichaevsky, Dr. Serg Ponomarev, Dr. Jung-Goo LeeGraduate students: Jessica A. Nash, H. S. Kim,
Nan K. Li, A. Singh, L. Sethaphong, R. C. Pani, A. Marlowe, H. Eksiri, M. Lim
Collaborators: A. Melechko (NCSU), Ralph Smith (NCSU); Stefan Zauscher (Duke University), Tosh Chilkoti (Duke University), Efie Kokkoli (U Minnesota)
Acknowledgements
